Professional Documents
Culture Documents
4-79 445
Printed in Great Britain
Swirling flow through a pipe is a highly complex turbulent flow and is still
challenging to predict. An experimental investigation is performed to obtain
systematic data about the flow and to understand its physics. A free-vortex-type
swirling flow is introduced in a long straight circular pipe. The swirling component
decays downstream as a result of wall friction. The velocity distributions are
continuously changing as they approach fully developed parallel flow. The swirl
intensity SZ, defined as a non-dimensional angular momentum flux, decays
exponentially. The decay coefficients, however, are not constant as conventionally
assumed, but depend on the swirl intensity. The wall shear stresses are measured by
a direct method and, except in a short inlet region, are a function only of the swirl
intensity and the Reynolds number. The velocity distributions and all Reynolds
stress components are measured at various axial positions in the pipe. The structure
of the tangential velocity profile is classified into three regions: core, annular and
wall regions. The core region is characterized by a forced vortex motion and the flow
is dependent upon the upstream conditions. In the annular region, the skewness of
the velocity vector is noticeable and highly anisotropic so that the turbulent
viscosity model does not work well here. The tangential velocity is expressed as a
sum of free and forced vortex motion. I n the wall region the skewness of the flow
becomes weak, and the wall law modified by the Monin-Oboukhov formula is
applicable. Data on the microscale and the spectrum are also presented and show
quite different turbulence structures in the core and the outer regions.
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
1. Introduction
At present much effort has been put into developing a more comprehensive
turbulence model that is applicable to complex turbulent flow. A t the Stanford
Conference on Complex Turbulent Flows (Klein, Cantwell & Lilley 198@1981), the
general consensus was that the models presented did not suffice for predicting
practical complex flows and needed improvement. Among the many factors which
make up a complex flow, streamline curvature and/or three-dimensionality are
important ones in the engineering fields. Many separate studies on the effects of these
factors have been made in the past. For example, Eskinazi & Yeh (1956), So (1975)
and Barlow & Johnston (1988) for curved flow, and Perry & Joubert (1965),
Bradshaw (1971) and Van den Berg (1982) for three-dimensional flow.
A swirling flow is a typical complex flow that is affected by streamline curvature
and flow skewness in addition to adverse and favourable pressure gradients. Their
combined effects on turbulence are quite unlikely to be described by a linear
16.2
446 0. Kitoh
combination of these factors. Thus, additional theoretical and experimental studies
are needed now for such a highly complex flow.
Kobayashi & Yoda (1987) attempted t o simulate numerically a swirling flow in a
pipe by using a k--s model and indicated that the velocity profiles obtained were
quite different from experimental ones. They concluded that the eddy viscosity
components were anisotropic, and that this was the main reason for the disagreement.
Computations based on a Reynolds stress model were performed by Gibson & Younis
(1986) for a swirling jet and by Hirai et al. (1987) for a swirling flow within two
concentric cylinders. These investigations met with some success in predicting the
flow. However, these flows were simplified versions of the swirling flow in a pipe,
because in one the wall effect was excluded and in the other had no core region. Lack
of systematic experimental information makes it difficult for any modeller to
compare his results with real data and thereby improve the model to predict the real,
physical processes in the swirling flow through a pipe.
The main objectives of this report are (a)to obtain systematic data including
velocity distributions, wall shear stresses and full Reynolds stress components from
up- to downstream sections and ( b ) t o discuss the velocity law and the turbulent
quantities in terms of hitherto known models in order to help explain the behaviour
of the swirling pipe flow.
2. Theoretical background
For engineering purposes it is important to understand the decay process of swirl
intensity along the pipe. Many workers, including Baker (1967) and Seno & Nagata
(1972), have reported an exponential decay formula for the swirl on the basis of the
empirical results.
For axisymmetric flow the swirl intensity Q, the non-dimensional angular
momentum flux, is defined as
where U , W and Urnare the mean axial, tangential and bulk velocity, respectively,
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
r and ro are radial position and pipe radius, and p is fluid density. This definition has
found acceptance with many investigators.
Because the swirl intensity decays owing to the tangential wall friction, we
consider it first. Integrating the tangential momentum equation written in cylindrical
coordinates ( r ,#, x), with # and x the polar and axial coordinates, respectively, we
obtain the tangential shear stress rrCas
rrC
ax
where V is mean radial velocity and the lower case (u,v,w) indicates fluctuating
velocities. The wall tangential shear stress normalized by the &Urn
becomes
k
&P,
= 2 d
lo
d(2/2r0)
‘oUWr2dr -
_ -1 d Q
r: Vrn 2 d(x/2ro)’
(3)
For extremely small 52, only the first term is important. Substituting this relation
into (3), the exponential decay formula can be obtained as
Q=Q,exp 2 a , d
{
whcre 0,and x, are the swirl intensity and the axial position of suitably selected
reference point, respectively. This formula is the same as that assumed by others. To
obtain the decay formula for large 0, we need an accurate relation between the
tangential shear stress and the swirl intensity, a formula that for the present can be
obtained only by experiments.
The velocity in a pipe is the axial flow upon which the swirling components are
superimposed. Thus the streamlines are spiral in the downstream direction and the
turbulent structures are subject to the mixed effects of centrifugal force due to the
streamline curvature and flow skewness caused by the non-uniform spiral pitch in the
flow. Both of these effects on the turbulent flow have been investigated separately by
many workers.
Perry & Joubert (1965) and Bradshaw (1971) pointed out that although the body
of the flow is significantly skewed there exists a thin shear layer close to the wall
where the flow does not skew much and obeys the ordinary law of the wall. Backshall
& Landis (1969) and Kit0 & Kato (1984) reported that the non-skewed layer very
close to the wall was observed experimentally in a swirling pipe flows. In this thin
layer, denoted as a wall region hereafter, the turbulence is affected by the streamline
curvature alone. Bradshaw (1969) indicated that there exists a strong similarity
between the effect of buoyancy and the streamline curvature, an effect which can be
estimated as the change of the mixing length, in analogy with the Monin-Oboukhov
formula developed for a stratified flow. By introducing the gradient Richardson
number Ri,thc mixing length 1 can be expressed as
1 = lo(1-pRi)
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
(6)
where p is a constant and 1, is the mixing length with no curvature effect. According
to So (1975) the gradient Richardson number for the curved flow should be expressed
as
+
where V, = (u2 W)) denotes the total velocity, y and R, are distance from the wall
and local radius of curvature of the streamline, respectively. In the thin wall layer
an inequality, y < R,,is also assumed. Close to the wall a steep velocity gradient
prevails and the relation aV,/ay % &/Reholds with sufficient accuracy. The mixing-
length formula then reduces to
where 7 is a total shear stress. Integrating the equation, the following velocity law
can be obtained : V = 1 YU*
A
u*
- In -+ B A V:
K v
+ (9)
448 0.Kitoh
where u*, K and B are a friction velocity based on the total wall friction, KSrman
and integration constants in the conventional log law, respectively. AV; denotes the
deviation of the velocity from the log law and is expressed as
where yo is the lower limit of the turbulent region and B is an additional integration
constant. Thus AV: changes linearly with s" YO
V,/u, dy/R, where - 2p is the constant
of proportionality.
Outside the wall region the skewness of the flow becomes noticeable and (6) no
longer holds. Some insight into the flow skewness in this layer can be given by
considering the transport equations of the Reynolds stress. In the skewed flow, the
flow direction 8, shear direction 8, and velocity gradient direction Og do not coincide
(Van den Berg 1982). The difference between 8, and Or can be estimated as
follows. The transport equation for ED and "W can be written according to Launder,
Reece & Rodi (1975) (referred to as the LRR model hereinafter). After boundary-
layer-like approximations according to Hirai, Takagi & Matsumoto (1988), these
equations become
where k and e are the kinetic energy of the turbulent motion and its dissipation rate,
and the constants in (11) and (12) are a = (8+Ci)/11, S = (8Ci-2)/11, y =
(30Ci - 2)/55, C, = 1.5 and Ci = 0.4. Diff qzjmeans the turbulent diffusion terms of
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
qzj.According to the numerical work on a swirling flow in coaxial pipes, with the
inner pipe rotating, the diffusion terms are negligibly small compared with other
terms (see Hirai et al. 1988). For simplicity, we also assume that the material
derivative terms D ( w ) / D t are small enough to be neglected. Then the shear stress
direction 8, = tan-l(vw/m) is approximated as
w- -
- (wz- v2)(2 -a) - S(u2- w2)
vw r
tan8, = - x tan8,+ ,
m au
-( -7+a;;"+ 8 2 - yk)
ar
where
Terms that include ED are neglected because of their smallness compared with other
normal stress components, see results from $6. Unlike the eddy viscosity model,
which predicts that the shear stress and velocity gradient are in the same direction,
this relation shows that 8, and 8, are in general different. Depending on the second
Turbulent swirling flow in a straight pipe 449
- _
term on the right-hand side, especially on the sign of aU/& and/or w2-v2, a complex
angle relation is expected.
The measured Reynolds normal stresses are convenient to use to check the quality
of the turbulence models. One of the most important but difficult points is the
modelling of a pressure-strain term which consists of the return-to-isotropy and the
rapid parts. Each includes an empirical constant, C , or C,, that should be determined
from appropriate experimental results. Adopting the LRR simplified model for the
- and applying similar approximations to the transport equations for 3
rapid-part,
and v 2 + w 2 as before, we obtain
0 = -2m--
au C
-
- (u,-$k) + 2C, m-au +;C,P -%,
ar 'k ar
where
Adopting an elaborate version for the rapid part in the LRR model, the ?/k relation
becomes
-
at
au
-= o ;
u2 2 2
- = -+-(a+/?- 1).
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
ar k 3 3C,
It is now easy to compare the above relations with the experimental data.
*
4
wer
Ventuii meter
FIGURE
1. Experimental apparatus. Dimensions in mm.
generator along which the flow passes radially inward. The vane angle can be fixed
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
by pins at 10' intervals from zero to 60" with respect to the radial direction. Six
different swirl intensities can be obtained. The bell-shaped cone at the centre of the
swirler smoothly deflects the radial flow in the axial direction (Yajnik & Subbaiah
1973). The flow rate can be adjusted by controlling the variable-speed motor. The
experimental Reynolds numbers Re based on the average bulk velocity Urnand pipe
diameter d are 40000, 50000, 60000 and 80000.
The measuring sections are prepared at the axial positions along the pipe shown
in table 1, at each of which there is a probe inserting hole. The measuring sections
(test sections) are denoted by TS in the figures and tables. Various types of hot-wire
probes are traversed radially through the hole to measure the mean velocities and the
Reynolds stresses. For measuring the flow angle 8 and the time mean velocity, an 1-
type hot wire made of tungsten (5 pm in diameter) having a sensing length of 1 mm
is used. Both end regions of the wire are copper plated (25 pm in diameter). As shown
later, the radial velocity is extremely small compared with the other components, so
the velocity vector is almost parallel to the wall. Using the flow angle 8 and the
resultant velocity V,, the axial and tangential velocity components U and W can be
easily obtained. Figure 2 shows the velocity components and the coordinate system
used.
Turbulent swirling $ow in a straight pipe 45 1
FIGURE
3. Types of x -wires used.
The complete set of Reynolds stress components are measured using the four
different types of x-wires shown in figure 3. They are similar in shape but have
different orientations, i.e. every 45", along the probe axis. The measuring principle
for the q is based on that of Muller (1982b). Fourteen sets of r.m.8. data of the
fluctuating velocity obtained per measurement- point from these probes, aimed in the
local flow direction, are enough to obtain all i&C, components. The tilde indicates
variables that are evaluated in the local coordinate system 2, y" and 2, where 2 is
directed to the flow direction and ij normal to the wall. The transformation from the
local to the laboratory system ( T , #, 2) can be made by simple algebra.
The nonlinear response of the hot wire to the oncoming velocity causes some error
in estimating the turbulent intensity. Muller (1982b) .indicated that neglecting
moments of the fluctuating velocity larger than the second induces some error when
452 0. Kitoh
_ _ _ _ _ ~ ~ ~ ______~
- - - -
~ ~
- -
x-wire type .liz 3 d2 Cfi 66 fid
I 0 0 0
I1 0 0 0
Ill 0
______~ O* o* o*
IV 0
TABLE2. x -type hot wires and Reynolds stresses that can be measured.
(* : Combination use of two probes.)
the turbulent intensity exceeds 20 %. Also, since the sensor has a finite size, another
error becomes appreciable when it is used in a region of steep velocity gradient, i.e.
in the central region and regions close to the wall in a swirling flow. To estimate the
accuracy of the measured Reynolds stresses, the redundant data from different sets
of measurements - and another measuring method are compared. Table 2 shows the
components of i&iij that can be measured by these probes. Figure 4 compares the
measured data from different x -wire types, and it includes results from rotating a
single wire (a method based on that of Bissonette & Mellor 1974).Consistency among
the data is very good except in the regions close to the wall and the pipe axis. Arrows
in the figure indicate the region of high turbulent intensity, larger than 20 or 10%.
Thus we conclude that the measured Reynolds stress data are reliable within
kO.001 UZ, except in the region of high turbulent intensity ( > 10%) and steep
velocity gradient. Hereafter the high-turbulence region ( >10%) is denoted by
broken lines.
For measuring the wall shear stress and the momentum fluxes, a test pipe of
50.8mm inner diameter and 4000mm long is used with the same type of swirl
generator but of reduced size placed at the upstream end. A three-hole cylindrical
Pitot tube, 2.5 mm in outer diameter, is used to measure the velocity components U
and W for evaluation of the non-dimensional angular momentum flux Q, the swirl
intensity. In a swirling flow the wall shear stress is difficult to measure by the indirect
method, which is based on the existence of the universal wall law. The wall law is
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
modified by the centrifugal force due to the rotating motion of the fluid, and one
cannot expect it to be a sound basis for indirect methods.
Here a shear-stress meter based on the direct method is developed. Figure 5 shows
the structure of the meter for the tangential wall shear stress T $ ~ The . sensing
element that is separated from the pipe system is suspended by two plate springs
so that it can be rotated by a small amount around the pipe axis if a torque is applied
to the element. The torque can be evaluated by measuring the rotation angle because
a linear relation exists between them for a limited torque range. The rotation angle
can be measured by the change of the distance between the target plate fixed on the
element and the electric eddy current probe. There are many error sources for a direct
shear stress meter as described by Allen (1977), and these sources are carefully
removed in our meter. The stiffness coefficient of the spring is selected so that the
torque resolution becomes as small as 8 x lo-' N m. The wall shear stress measured
by this meter is the average over the element surface, which is 24mm in length
(0.47d). The evaluated mean value does not coincide exactly with the local wall shear
stress, because in general the swirling flow is axi-asymmetrical (Kito 1984), and the
shear stress is distributed around the periphery of the pipe. To check the final
accuracy of this meter, the well-defined laminar flow between two concentric
Turbulent swirling flow in a straight pipe 453
0 0.5 1.o
TITO
0.01 1 I 1 I
I
1
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
rIr0
FIQURE 4. Accuracy of Reynolds stresses measured by x -wires. x -wire type : A,I; 0 , I I ; 0 , 1 1 1 ,
IV. -, Single rotating wire (Bissonette & Mellor 1974). Regions of turbulent intensity larger
than 20% (-)) and 10% (---D). Error bars indicate an uncertainty band of *O.OOIVm.
cylinders, the inner one rotating, is used. The measured torque coefficients coincide
with the theoretical ones to within an accuracy of f3.5 YOof the reading (20 : 1 odds).
For measuring the axial wall shear stress 7 z w ,the same meter is used, except that
eight music wires are used instead of the spring. The force resolution is 1.5 x N.
The accuracy of the data is checked by measuring T,, in fully developed pipe flow
and comparing with the Blasius or Nikradse formulae. The accuracy is estimated to
be f4 YOof the reading (20 :1 odds).
454 0. Kitoh
I 1
FIGURE
5. Wall-shear-stress meter: a, plate spring; b, sensing element of pipe; c, target plate;
d , electric eddy current probe.
in equation (5). Sen0 & Nagata (1972) considered the effect of the pipe roughness and
Baker (1967) and Padmanabhan & Janek (1980) studied the effect of the Reynolds
number and inlet swirl intensity on the decay.
Figure 6 shows the decay of D plotted on a semi-log map at Re = 50000 for various
inlet swirl intensities, SZ being evaluated from the Pitot tube measurements. The
abscissa x’ is defined as the distance from the virtual origin x,. The x, are selected so
that each decay curve for a different inlet swirl intensity would fall along a single
curve. The decay curve is approximately linear but has a ‘knee’ point a t Q = 0.1.
Thus the decay coefficient 2a, in (5) is expected to depend on the swirl intensity. A
more detailed picture of the decay coefficient can be obtained from the relation
between the tangential wall shear stress T+,, and Q. Figure 7 shows this relation, the
former measured by the shear-stress meter, for various inlet swirl intensities. To
obtain the data the shear-stress meter was successively moved from xld = 6.3 t o 80.3
in increments of 5d. The data fall along a single curve, denoted by the solid line in the
figure, irrespective of the different inlet swirl intensities except in the inlet region, the
portions denoted by broken lines. The arrows on the line show the downstream
direction. Thus T~~ can be expressed as a function of 52 except in the inlet region. The
solid line has clear deflection points at D = 0.09,0.45and a less clear one a t 52 = 0.04.
Turbulent swirling flow in a straight pipe 455
1 L I I I I I I I I
I 1 I I I
-
I - '
- -
-
a
0.1 r
-_
- %;; --
\
- =\ I
0.01 I I I I I I I 1 I I I I I I
FIQURE 6. Decay of swirl intensity SZ along pipe axis at Re = 50000, x' is distance from the
virtual origin 2,. Swirl intensity and distance at initial reference section (x/d = 23.4); 0 , 8= 0.6,
x,/d=O; A, a = 0 . 2 7 7 , x,/d=32; .,SZ=0.256,x0/d=36.1; A,SZ=0.173, x,/d=52.6; 0 ,
G! = 0.102, x,/d = 70.2; a,a
= 0.081, x,/d = 82.6.
a
0 0.05 0.10 0.15
0.010
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
0
0 0.5 1 .o
a
FIQIJRE 7. Relation between tangential wall shear stress and SZ. Initial swirl intensity estimated
a t x / d = 6 . 3 : Q), Q=0.966; @,0.631;0 , 0.431; a, 0.279; 0,0, 0.178; 0,0.130; .,0.086.
456 0. Kitoh
3 -
- 2 A i x 100 -
I 1 I I
1 2 5 10 15
Re x 10-4
FIGURE
8. Decay coefficient of a : 0 , -2A,; 0 , -2-4,; A, -%0I
,, ;
-2-4,; -,
Baker’s (1967) result.
The change of the slope at 52 = 0.09 corresponds to the knee point in figure 7, but
those at 0.45 and 0.04 have no clear corresponding points on the swirl decay curve.
If we assume piecewise linear relations between the knee points, r4, can be expressed
r4,/$pPm= A i Q + B i , i = 1 , . .. , 4 .
as
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
The constants A , and B, are tabulated in table 3. Using the linear relation of T+, with
Q, the decay formula can be expressed as
log@+:) = 2At7+log
x-xr (Qr+- 3.
The exponential decay law is correct only in the range 0 < 52 < 0.04, because B, = 0
here. When B, < A , , the exponential decay law holds approximately and 24, is
nearly equal to a decay coefficient, 2a1.Figure 8 compares the coefficients 2A, to the
decay coefficient reproduced from Baker’s (1967) diagram. Basically, it decreases
with Re. The difference of 2A, for various ranges of 52 is significant; in the range
0.09 < 52 < 0.45, 24, is especially small compared with those for different Q.
The variations of the axial wall shear stress r,, with 51 for three Re values are
shown in figure 9. r,, increases noticeably with 52, and is almost double the zero-swirl
value at 52 = 0.55. The Reynolds-number dependence of the wall shear stresses can
be estimated by considering the variation of a friction factor of the pipe, A, with Re
like
Turbulent swirling jlow in a straight pipe 457
0.015
I
0
0
0
0 0
0
00
0.005
0 0.5 1.o
n
FIGURE9. Relation between axial wall shear stress and 0 : 0 ,Re = 50000;0 ,Re = 100000;
A, Re = 150000.
1 1 1
I I I 1 1 '
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
0.5 1.o
D
FIGURE 10. Shear-stress direction at the wall measured by the shear-stressmeter. 0 ,Re = 5OOOO;
-, Re = 100000;-----, Re = 150000; 0 ,streamline direction at the wall measured by a hot-
wire anemometer.
458 0.Kitoh
which is within an accuracy of + 2 YOin the range of 5 x lo4 < Re < 15 x lo4. A
similar relation holds for tangential shear stress to within f5 YOaccuracy. The shear-
stress direction 8, a t the wall changes with 52 and the relations measured by the
shear-stress meter are shown in figure 10. The limiting streamline angles with respect
to the pipe axis measured by a hot-wire anemometer are also included in the figure.
The overall agreement between the two methods is good. Some scatter of the data
from the hot wire is seen a t around 52 = 0.2. Kit0 (1984)reported that axisymmetry
of the velocity profile fails near 52 = 0.2, so that the local streamline direction a t the
wall also becomes asymmetrical with respect t o the pipe axis, and this appears to be
the main reason for the scatter of the data from a hot wire.
The wall shear stress data 74w and 7,, presented here are average values over the
pipe circumference and would be appropriate for comparing with the results of a
numerical simulation of swirling flow, for in numerical work it is usually assumed
that the time-mean values are axisymmetrical.
5. Velocity distributions
Figure 11 ( a , b ) shows the profiles of tangential and axial velocity distributions W ,
U at various stations along the pipe axis as measured by a hot-wire anemometer.
From the shape of the tangential velocities the swirling flow has a three-region
structure consisting of wall, annular and core regions. I n the wall region the velocity
gradient is quite steep and, as described later, the flow skewness here is negligibly
small. The annular and core regions are characterized by free- and forced-vortex-
type velocity distributions, respectively. I n the figure the distribution of a pure
forced-free vortex model (Rankin’s combined vortex) is shown for comparison.
Except in the region around the meeting point of two vortices, the Rankin vortex is
a reasonable qualitative model of the tangential velocity distribution unless 52 is too
low. The extent of each vortex region changes in accordance with the swirl intensity.
In a weak swirl, 52 less than 0.2, the boundary between the core and annular regions
becomes obscure, and finally, when 52 < 0.1, the forced-vortex-type motion
dominates the whole section.
Pressure within the section can be estimated from
}
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
w
P = P w - p r { T - - - (l a
r ar
-
rv2) d r
where 3 < W is assumed. Using the turbulence data to be given later, the effect of
normal stress on the pressure distribution is less than 3%. The wall pressure P, is
measured as a mean value from four pressure holes that are equally spaced around
the circumference. Figure 12 shows the pressure distributions across the section.
Owing t o the swirl decay along the pipe axis, the centrifugal force weakens and
adverse and favourable pressure gradients appear in the centre and outer regions,
respectively. These pressure gradients have a considerable effect on the axial velocity
distributions, i.e. a low velocity in the centre region surrounded by high velocity in
the annular region, figure 11 ( b ) .When the swirl intensity is larger than 0.4, in our
experiment, the reverse flow appears a t the centre and its magnitude and the area
increase as 52 increases.
The radial velocity V is estimated from a continuity equation
Turbulent swirling flow-in a straight pipe 459
2.0
1.5
1 .o
I
1.5 I
I
1
0.5
I .o
W
- U
-
u,
0
0.5
I
I
I
-0.5#
0
I
-1.0 I I I I
I 0.2 0.4 0.6 0.8 0
rlr,
0.0025
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
-
V
- 0.0025
0 0.5 1 .o
rlr,
FIQURE 11. (a) Tangential, ( b ) axial and (c) radial velocity distributions. Run 7 : A, TS 3. Run 8:
0, TS 1; 0, TS 3 ; 8 ,TS 5; (>, TS 7 ; 0, TS 9; 0 , TS 10. Run 10: 0, TS 7 ; 0 , TS 10. ----.
Rankin combined vortex. (TS and runs shown in table 1 . )
460 0.Kitoh
P- P,
__
&XJ;
-2
c
0 0.5 1.o
rIr0
12. Pressure distribution across a section. P, is a reference pressure (=P,(x/d= 41.1)).
FIGURE
For symbols, see figure 11.
and the distributions of V in a cross-section are shown in figure 11 (c). The magnitude
of V is of order 1/1000 of the average axial velocity Urn,and the total velocity
direction is considered to be almost parallel to the wall surface.
5.1. Wall region
Figure 13 examines the skewness of the flow close to the wall by plotting U versus
W . Up to some point from the wall the flow angle does not change, collateral flow,
and it then changes gradually as the wall distance increases. For a three-dimensional
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
flow developed on a flat plate, Johnston (1970) and Perry & Joubert (1965) reported
that the edge of the constant-flow-angle region is at 6CL100, or 150 in the wall
variable y+. Here this edge point is situated at around 60. The centrifugal force is the
dominant term that affects the flow within this constant-angle region. Figure 14 (a+)
shows semi-log plots of V, for ( a )weak, ( b ) moderate and (c) strong swirl intensity and
in comparison with the conventional log law. The friction velocity is estimated from
the velocity gradient at the wall. In the range 0 < D < 0.11 the velocities follow the
conventional log law up toy+ x 200 and the streamline curvature has no effect on the
flow. For swirl intensity larger than SZ > 0.3, the data (y' < 1000) fall along a single
curve with a smaller slope than the conventional curve, irrespective of the swirl
intensity. This means that the centrifugal force effect on the turbulence is saturated
when SZ exceeds 0.3. I n the intermediate swirl intensity range, the slope changes
gradually with Q. Thus, within the rather narrow range of D from 0.1 to 0.3, the
centrifugal force effect on the turbulence starts and reaches saturation.
The deviation of the velocity from the log law AV; can be expressed by equation
(10). The excess value of the velocity AV; is plotted in figure 15 against l/R,soy
soy
(V,/u,)dy. Here the lower limit of the integration is zero instead of yo, which results in
only a small change in the abscissa. In the range 0.02 < 1/R, (V,/u,)dy < 0.1 the
Turbulent swirling $ow in a straight pipe 46 1
0 0.5
w/urn
13. Flow skewness in the near-wall region. Run 1 : 0 ,TS 6;8 ,TS 9. Run 2 : 0 ,TS 9.
FIGURE
Run 3: V,TS 9.
data for AV; fall on a line of slope - 12. The furthest extent of the validity of (lo),
i.e. the wall region, is thus considered to be the y-coordinate that gives an integral
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
value of 0.1. According to (lo), this result shows that the constant /3 = 6. The
constant in the Monin-Oboukhov formula has been reported by many workers for a
flow passing through a curved or rotating channel and it is distributed in the range
2 < /3 < 7 for small Richardson number. An overall review for /3 was given by So
(1 975) and he recommended /3 = 2.5 from among the many previously proposed
values. The result obtained here is not consistent with So but it is the same as that
obtained by Johnston, Halleen 6 Lezius (1972) for rotating channel flow.
The slope of the log formula in the semi-log diagram, equal to the inverse of the
Karman constant 1 / ~ changes
, as the streamline curvature effect appears. Here the
change of K with B can be used as a measure of the curvature effect on the flow in
the wall region (figure 16). From S2 = 0.1 to around 0.3-0.4 it increases from the
known value of 0.41 to the final value of 0.725.
The change of the Richardson number with S2 is interesting, for it also expresses
quantitatively the curvature effect. The figure also shows the change of -Ri
estimated at y+ = 100 with Q. The magnitude of Ri increases from 0.05 at Q = 0.1
to a final value of 0.27 at S2 = 0.4-0.5. The curve is very similar with that of K . The
curvature effect is almost saturated a t around SE = 0.4-0.5. Although there exists
some scatter in the data points, -Ri at S2 = 0.1 is about 0.05. Townsend (1976)
462 0.Kitoh
30
20
0
10 100 1000
Y'
FIQURE 14. Log law in swirling flow, (a)weak swirl, ( b ) moderate swirl, (c) strong swirl. -,
~ / u , = 2 . 5 1 n y C + 5 . 5 . R u n 1O
: ,TS2;O,TS6;e,TS9;@,TS 10.Run2: e , T S 4 ; O , T S 6 ; 0 ,
TS9; u,TS 10; 0 , TS 11. Run 3 : A,TS6; 'I, TS 10; m, TS 11. Run 4 : m, TS 1 1 . Run 5 : 0,
TS 11. Run 6 : 6 , TS 9.
1
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
-
AVt
I(*
-1
I I I I I
-
0 0.1 0.2 0.3
15. Deviation of mean total velocity from log formula, AV;. -, Line of slope - 12.
FIQURE
For symbols, see figure 14.
Turbulent swirling flow in a straight pipe 463
1"'""""'" 0 0
15
-Ri
1
35
U U. 3 1 .u l.J
R
FIGURE 16. Effects of swirl intensity 52 on the KBrmLn constant K , Richardson number Ri and
radius of streamline curvature R,: -0-, K ; ---n----, Ri at y+ = 100; ----, R,;
, above this Ri line, the effect of centrifugal force on the flow becomes appreciable,
Townsend (1976).
suggested that the curvature effect becomes appreciable when pi1 2 0.05. This
suggestion is consistent with the present result that the centrifugal force becomes
important if l2 exceeds 0.1. Such a rapid change of K or Ri from l2 = 0.1 to 0.5 and
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
its almost levelling off beyond 0.5 can be explained by the change of radius of
streamline curvature with 52. The radius of streamline curvature R, at the wall can
be estimated as @/(sinz8 ) .Using the data for 8 in figure 10, the R, curve in figure 16
w..-l
a . n ent,ima.t.ed
-I --
I-___-I--. ---___
From .O= 0 1 t,n n
"._ I-
2 -"c
".-, R denrc?a.nen- - ~ - - -a,a
--------- ra.nidlv -~ --
I
B ----------
innreasen hiit,
--- its
7---
reduction rate becomes smaller. Beyond 52 = 0.5, R, does not change so much and
the curvature effect is almost saturated there.
5.2. Annular region
Figure 17 shows the distributions of 0,0, and 8, (i.e. flow, velocity gradient and shear
stress angles, respectively). The hatched region in the figure indicates the wall region
and the three angles coincide there. Beyond the wall region in the annular region,
different angle relations become noticeable : OSand 0, change rapidly along r, whereas
changes in 8 are slower. Thus, in this region the flow and shear directions skew and
the three angles do not coincide with each other. The turbulent motion contains the
skewness effect in addition to the curvature effect. We cannot expect the simple
mixing-length model of turbulence to work well here.
In this region the tangential velocity approximates a free-vortex-type distribution.
464 0.Kitoh
150
50
0 0.5 1.o
TITO
FIQURE 17. Relations among angles 8, 8, and 8,. R u e 8_TS 7 : 0 , 8 ; 0 , 8,; A, 8,; ----, radial
position of aU/ar = 0 ; radial position of w2-va = 0. Shaded region is the wall region.
As the swirl decays downstream, the velocity distribution changes from a free vortex
to a forced vortex type. For a nearly plane swirling flow (the axial gradient is very
small compared to the radial gradient) Reynolds (1961) indicated that a free-vortex-
type flow is expected from an order estimation of the momentum equation. This
estimation was substantiated by the work of Hirai et al. (1987)who measured the free
vortex motion in a plane swirling flow within two concentric cylinders, the inner one
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
rotating and the axial velocity component being superimposed on it. In the swirling
flow decaying downstream, however, the effect of axial transport of momentum
becomes important as shown later and the momentum balance among many terms
becomes more complicated than in the plane swirling flow case. In this case we
cannot say anything about the velocity distributions from a theoretical standpoint.
From a phenomenological point of view, we assume that the tangential velocity in
the decaying swirling flow can be expressed as a modification of the plane swirling
flow like
w -+f(k),
c
urn = r l r o
-
I I I I ~ I I I I 1
- -
/a
1.0 - 1-
O/
/
/
-
/
7/
d -
'9
c, w
- m '
/
0.5. - 9/6 -
/
0 /o -
0 B.' a
-
/ 0
- a9 -'
"0'
00 Q
p,% 0 a -
@'I I 1 I I I I I I 1 L
D decreases.
According to the Rayleigh criterion (dldr)( r W ) 2> 0 for stability in relation to
small perturbations, the velocity profile of this type has a strong stabilizing effect on
the turbulence and the small-scale turbulent motion should die out rapidly. This
situation is confirmed in the next section. Thus, in the core region large-length- and
timescale motions prevail. In this region it is expected that the large-scale motion
could persist for a long distance downstream and indicates some history effect on the
flow.
Figure 19 (a, b) compares the velocity distributions U and W which have nearly the
same local swirl intensity but different upstream conditions, i.e. the flows were
generated by different swirlers and have different inlet swirl intensities and axial
lengths. For example, Weske & Sturov (1974) adopted a special swirler that
generated pure forced vortex motion by rotating a grid. The central part of the
velocity differs significantly for each case, whereas the profiles in the annular region
are similar. These results indicate the long history effects in the core region.
466 0.Kitoh
FIQURE
19. Velocity distributions (a)W and (b) U having nearly the same swirl intensity but
different upstream conditions. @, Present, Run 1, TS8; +, Sen0 & Nagata (1972), Q = 0.73,
Re = 19 x lo4,x/d = 19; x , Weske & Sturov (1974), Q = 0.68, Re = 3 x lo4, x/d = 20.
6. Turbulence
6. measurements
Turbulence measurements
Figure 20(a-f)
Figure 20 (u-f) shows
shows the the successive
successive changes
changes of of the
the normal
normal stresses 22
stresses and and their
their
production P($)from the upstream to the downstream direction measured by the
production P ( 2 )from the upstream to the downstream direction measured by the
x -wire system. In the figure the data of Weske & Sturov
x -wire system. In the figure the data of Weske & Sturov (1974)for swirling flow and(1974) for swirling flow and
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
Figure 24(a-c) shows the eddy viscosity distributions across the section. The data
scatter widely, but show some relation with r. The large anisotropy among the three
components can be seen, ahd this result is consistent with Lilley & Chigier's (1971)
report on the anisotropic nature of turbulent viscosity in swirling flow. Very close to
the wall the anisotropy becomes weak. Kobayashi & Yoda (1987) adopted k--s
model to simulate swirling flotk and concluded that the standard model did not
predict the velocity distribution well. However, they could obtain a satisfactory
result when an anisotropy factor at, was introduced in the relation vtij = aijCk+/-s.
- Figure 25 indicates the ratio of shear stress to twice the turbulent kinetic energy
q2,a, = (m2 +m2)i/?, for various swirl intensities. For a two-dimensional flow
Bradshaw, Ferris & Atwell (1967) obtained good numerical results by putting a, =
0.15. Recently Muller ( 1 9 8 2 ~reported
) that even in a skewed boundary layer the
468 0.Kitoh
0.2
Q,L
0.1
0
11
r
--
,111
0.5
1111(
1.o 0.5 .o
1
r/ro r/r0
*3'0
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
1 1 1 1
0 0.5 1.o 0 0.2 0.4 0.6 0.8 1.0
r/r0 r/r0
FIGURE
20(a-d). For caption see facing page.
Turbulent swirling jlow in a straight pipe 469
6
I I I I
1I
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
rlrQ rlr0
FIGURE
20 (a-f). Distributions of Reynolds normal stresses and their production. For symbols,
see figure 11. ---, Laufer (1952);----, Weske & Sturov (1974).
Parallel flow
-2m-
au
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
ar
-m-au
ar
--87
v2 -
ar
LRR model
GY
Simplified Elaborate model
-
- Experiment
UZ
- at aU/ar=O 0.684 0.626 0.489 0.609 0.511
- _
k
(Run 8. TS 1 - TS 10)
v2+w2
at awl&- W/r =0 1.098
k (Run 10. TS 7)
1.078 0.978 1.022
(Run 10, TS 10)
TABLE
5. Comparison of q / k predicted from Reynolds stress models and experiment.
factor a, is constant a t 0.15. For a two-dimensional curved flow, however, Gills &
Johnston (1983) and Barlow & Johnston (1988) indicated that the factor ai is not
constant through the boundary layer, with a trend that depends on whether the
surface is convex or concave. On the concave side, unstable flow, a, increases over
40% above 0.15, while on the convex side it decreases to under 0.15. In swirling flow
aiis far from constant ; first it increases to 0.15 a t r/ro = 0.9 from zero a t the wall
then decreases to 0.09 in the annular region. According t o the Rayleigh criterion, the
tangential velocity distribution in the annular and core regions is stable as in the case
of a boundary layer on a convex wall, Gills & Johnston (1983). The low value of ai
in these regions could be attributed to the centrifugal stability effect, but clear
evidence for this conclusion could not be given here. I n the core region ( r / r , < 0.4)
the value decreases further.
In the annular region of highly skewed flow, the shear and velocity-gradient angles
differ from each other. According t o equation (13) the relation betwccn _ _ the two
angles is dominated by the sign of the leading terms of aU/ar and ( w 2 - v 2 ) in the
annular region. Figure 17 shows the cross-sectional variation of Og and 0,. I n the
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
annular _ _ the curves of 0, and BS cross a t two points. I n the figure the radial
region
position w2- v2 = 0 is indicated by a chain dot line which corresponds to the crossing
point near the wall. Figure 26 compares the measured angle relation with the
Reynolds stress model, equation (13), whcre the measured turbulence data are used
in the model. The qualitative agreement is good.
The turbulent character is expected to be different among the core, annular and
wall regions because of the centrifugal stabilizing or destabilizing effect in the
swirling flow. The W-distribution here satisfies the Rayleigh criterion for stability
throughout the cross-section except close to the wall but we do not have any evidence
of suppression of the fluctuating motion. Eskinazi & Yeh (1956) performed a two-
dimensional curved flow experiment and their data indicated that the stable region
estimated from the Rayleigh criterion did not exactly coincide with the area where
the turbulent kinetic energy was absorbed by the mean velocity. I n addition to the
stabilizing/destabilizing effect of the tangential velocity, there is another energy
source of the turbulence, production of 2,in the swirling flow. This additional energy
source makes an important contribution t o the energy balance, making the situation
more complex than two-dimensional curved flow.
I n the core region, because of the relation d(rW)2/dr>> 0, the stabilizing effect
Turbulent swirling flow in a straight pipe 471
-
-0
,
c
c
0
-I
0.2 0.4
TITO
0.6 0.8
4
1.0
-0.01
0 0.2 0.4
rlr0
0.6 0.8 1.0
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
-0.005 I 1 I I 1 I
0 0.2 0.4 0.6 0.8 1.0
rlr0
t
t 1
(b)
- 0.005
0 0.5 I .o
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
TITO
FIGURE
22. (a) Balance of axial momentum flux across a section. Run 8 : TS 7.
Figure 27 (a, b ) shows the time trace of the velocity fluctuation and its spectrum
measured in the wall, annular and core regions. In the core region very low-frequency
motion prevails, while in the outer regions the fluctuation includes high-frequency
motion as expected in turbulent flow. The energy spectrum of the streamwise
velocity also reflects this tendency. I n the core region a peculiar frequency is
observed and this might be the signal of an inertial wave generated by the rotating
Turbulent swirling flow in a straight pipe 473
0.10
t 1
0.05
-0.05
-0.10
motion. The microscale of the turbulence A,, estimated on the assumption of the
frozen turbulence and using the relation
is shown in figure 28. It becomes very large in the core region when l2 is large, thus
indicating the non-dissipative character of the motion here.
474 0.Kitoh
0.010
0.005
&
urn ro
I I
0
I I I I
\
0
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
FIGURE
24. Distributions of eddy viscosities, Re = 50000.
7 . Concluding remarks
The tangential velocity in swirling flow through a pipe has a significant influence
on the flow structure. Among many factors that affect the flow, we limited
consideration to thc ccntrifugal stabilizing/destabilizing effect, flow skewness and
history effects. Based on the tangential velocity distribution, the flow has three
regions : wall, annular and core. In the wall region only the centrifugal destabilizing
effect appears and the classical mixing-length model modified by the Monin-
Oboukhov formula can predict the flow there. The coefficient /3 is determined to
Turbulent swirling flow in a straight pipe 475
I , # I '
- - - - - - - - - - - -Eta-
0-
0.15
0
O 00-
a,
:-
0
88 P
0.10 -
G
e0
:"
8
A
0
(D
..t%g"
0.05 I a L O l ' I B ' I I
0 0.5 1.o
, "
FIGURE ~ ) p.
25. Ratio of shear stress ( E V * + E W to ~ For symbols, see figure 11. ----, a, = 0.15.
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
tan 0, -
FIGURE
FIGURE
26. Comparison
26. Comparison of relation between
of relation between angles 0, and
angles 0, 8, between
and 8, between experiment
experiment and
and Reynolds
Reynolds
stress model.
stress model. Run
Run 8,8, TS 0,,experiment
TS 77:: 0 experiment;; xx ,, Reynolds
Reynolds stress
stress model.
model.
16
16 FLM 225
FLM 225
476 0.Kitoh
I I 1 I I I
0 12
1 1 1 I I I
(ii)
'C I I I I
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
1
-1
0 4 8 12
tl(r01 Urn)
be 6 to fit the experimental data that is consistent with Johnston's (1970)result. The
annular region is characterized by a flow skewness and angles 9, 9, and 9, that are
different from one another. In this region the analytical approach to predicting the
flow is quite difficult. A Reynolds stress model which can handle anisotropic
turbulence is a more promising means to model the flow compared to an eddy
viscosity model like k--E. The angle relation between 9, and 9, based on this model
$ow in a straight pipe
Turbulent swirling flow 477
0 1 2 3
log (krJ
FIQURE 27. (a)
(a)Time records of velocity fluctuation 4 at three radial positions: (i) r / r O = 0.067 ; (ii)
Run 7, TS 3. t in abscissa indicates time. ( b ) Energy spectrum of C2.
r/rO= 0.733, (iii)r/ro = 0.973. Run C2.
Run 7, TS 3. kk in abscissa indicates wavenumber. 0 , r/ro = 0.067
0.067; A, r / r o = 0.733; 0 , =
0.973.
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press
\
\
- \ -
\
\
.\I 0.1-9
\
' -
18-2
478 0.Kitoh
is qualitatively in agreement with the experiment. It is also noticeable that the
tangential velocity could be expressed as a sum of forced and free vortex motions. In
the core region, the centrifugal stabilizing effect becomes important, turbulent
motion with very low-frequency prevails and the flow is non-dissipative. The inlet
conditions have a large effect on the downstream flow in the core region. Thus for the
core region there is a long history effect.
REFERENCES
ALLEN,J. M. 1977 Experimental study of error sources in skin-friction balance measurements.
Trans. ASME I : J . Fluids Engng 99,197.
BACKSHALL, R. G. & LANDIS, F. 1969 The boundary-layer velocity distribution in turbulent
swirling pipe flow. Trans. ASME D : J . Basic Engng 91, 728.
BAKER,D. W. 1967 Decay of swirling, turbulent flow of incompressible fluids in long pipes.
Dissertation, University of Maryland.
BARLOW, R. S. & JOHNSTON, J. P. 1988 Structure of a turbulent boundary layer on a concave
surface. J. Fluid Mech. 199, 137.
BISSONETTE, L. R. & MELLOR, G. L. 1974 Experiments on the behaviour of an axisymmetric
turbulent boundary layer with sudden circumferential strain. J . Fluid Mech. 63, 369.
BRADSHAW, P. 1969 The analogy between streamline curvature and buoyancy in turbulent shear
flow. J . Fluid Mech. 36, 177.
BRADSHAW, P. 1971 Calculation of three-dimensional turbulent boundary layers. J . Fluid Mech.
46, 417.
BRADSHAW, P., FERRIS, D. H. & ATWELL, N. P. 1967 Calculation of boundary-layer development
using the turbulent energy equation. J . Fluid Mech. 28, 593.
ESKINAZI,S. & YEH, H. 1956 An investigation on fully developed turbulent flows in a curved
channel. J. Aero Sci. 23, 23.
GIBSON,M. M. & YOUNIS,B. A. 1986 Calculation of swirling flow jets with a Reynolds stress
closure. Phys. Fluids 29, 38.
GILLS,J. C. & JOHNSTON, L. P. 1983 Turbulent boundary-layer flow and structure on a convex
wall and its redevelopment on a flat wall. J . Fluid Mech. 135, 123.
HIRAI,S., TAKAGI, T. & MATSUMOTO,M. 1988 Prediction of the laminarization phenomena in an
axially rotating pipe flow. Trans. ASME I : J . Fluids Enqnq 110, 424.
HIRAI,S., TAKAGI, T., TANAKA, K. & KIDA,K . 1987 Effect of swirl on the turbulent transport of
momentum in a concentric annular with a rotating cylinder. Trans. JSME B 53, 432.
https://doi.org/10.1017/S0022112091002124 Published online by Cambridge University Press